The first time I backtested a naive congressional-disclosure strategy, the results looked too good, and that was the tell. Buy what the well-connected buy, hold for a couple of months, and the equity curve slopes up in a way that makes you want to trust it. What the clean backtest hides is that the edge you are measuring was mostly harvested in an era when almost nobody was watching. Once trackers, ETFs, and a few loud social accounts turned those filings into a spectacle, the drift they were feeding on got a lot thinner. The strategy did not stop working overnight. It just quietly stopped paying what the backtest promised.
This is the crowding problem, and disclosure trading is one of the cleaner places to see it, because the signal is public, timestamped, and easy to copy. When a signal is easy to copy, it gets copied, and the excess return it used to carry gets competed away. The interesting question is not whether decay is happening. It is how you measure it in your own signal, and what you do once you accept that reaction speed now matters more than the thesis.
How the drift gets compressed
The whole premise of a disclosure strategy is post-disclosure drift. A filing lands, the information is technically out, and yet the price keeps moving in the disclosed direction for days or weeks afterward because attention diffuses slowly. That drift is the alpha. It exists because the market is inefficient about processing a firehose of scattered filings.
Crowding attacks the drift from both ends. On the front end, faster participants front-run the slow diffusion. If a hundred bots and three ETFs all read the same filing within minutes, the price adjustment that used to take a week now happens in a day, sometimes in an hour. On the back end, the crowd changes the character of the disclosed trades themselves. When people know they are being watched and mirrored, some of them get more careful, some get performative, and the average informativeness of the signal drops. You end up with more copies chasing a weaker underlying edge, which is the worst of both.
The practical consequence is that the shape of your return moves earlier and gets shorter. The gains that used to sit in days two through fifteen migrate toward hours zero through two, and the tail flattens. If your execution assumes you can enter calmly the morning after a filing, you are now entering into a price that has already done most of the work.
Measuring decay in your own signal
You cannot manage what you refuse to measure, and most people running these strategies never actually check whether their edge is eroding. They look at cumulative return and feel fine. The number that matters is not the total, it is how the per-trade excess return behaves over time and how much of it survives realistic execution.
Here is a workflow I keep coming back to. It is not fancy, and that is the point.
- Anchor every trade to the disclosure timestamp, not to the trade date the filer chose. The clock that matters is when the information became public and copyable.
- Build an event-time return curve. Average the abnormal return across your trades at hour one, hour four, day one, day three, day ten, and so on, measured against a sensible benchmark so you are isolating the signal and not just riding the broad market.
- Cut that same curve across time periods. Roughly the last few years versus the few before that. If the curve is flattening and shifting left, you are watching your drift get eaten.
- Overlay a realistic entry. Re-run the whole thing assuming you enter with a delay, with slippage, and with the spread you actually pay. The gap between the paper curve and the executable curve is the part crowding has already taken from you.
- Track the half-life. Estimate how long it takes for half the eventual drift to show up. When that half-life keeps shrinking quarter over quarter, speed has become the whole game and your comfortable next-morning entry is dead.
The failure mode I see most often is survivorship dressed up as skill. People remember the two trades that ran for a month and forget the fifty that were flat by lunch. Event-time curves fix that because they force every trade into the same frame and let the average tell you the truth. If you only ever look at winners, you will conclude the strategy is fine right up until the drawdown that convinces you it never was.
When speed becomes the strategy
Once the half-life is measured in hours, the thesis stops being the differentiator. Everyone reading the same filing has roughly the same thesis. What separates a real edge from a copy of a copy is how fast and how cleanly you act on it, and whether your costs let you keep any of the move.
A few things I would hold as rules of thumb once you accept a signal is crowded:
- If most of the drift now happens before you can realistically trade, the honest move is to size down or drop the signal, not to lever up on the leftovers.
- Latency from filing to your order is a real cost line. If you are learning about a disclosure hours after a tracker already tweeted it, you are the exit liquidity, not the edge.
- Crowded signals get more correlated with each other, so a portfolio of them is less diversified than it looks. Ten copies of the same trade is one trade.
- Turnover kills. As the payoff window shrinks, you trade more often for smaller edges, and fees plus slippage quietly become the dominant term. A signal that looks positive gross can be flat or negative net.
Where there is still room
The good news is that crowding is uneven. The signals that got famous are the ones that got crowded, and fame in this world is narrow. Congressional trades and a handful of high-profile insiders soaked up the attention. Plenty of disclosure sets sit adjacent to them, carry similar information, and have almost nobody watching.
Executive-branch and agency-level disclosures are a good example. The filings exist, they are public, and they are far less packaged into tidy trackers than the legislative stuff, so the diffusion is still slow and the drift still has a tail. International filings are another. A company cross-listed under two disclosure regimes has to answer to both, and those regimes run on different clocks and different formats, which means the same insider event can surface in one jurisdiction before the crowd in the other has noticed. Less liquid names help too, not because the signal is stronger but because fewer machines bother to scrape them, so the attention that compresses drift never arrives.
The tradeoff is real and worth saying plainly. Less-crowded sets are less crowded partly because they are annoying. The data is messier, the filings are harder to parse, the names are thinner, and you cannot deploy much capital before you move the price yourself. That friction is the moat. If it were easy to industrialize, it would already be crowded, and the drift would already be gone. The work of normalizing scattered filings into one clean, timestamped stream is most of the edge, which is a chunk of what we spend our time on at Blockcircle.
None of this means disclosure trading is dead. It means the free version is, the one where you copy the famous trades a day late and expect the old numbers. Measure the decay in your own signal honestly, respect the clock, and spend your effort where the crowd has not shown up yet. That is a less exciting story than the backtest told you, and it is the one that still holds up when real money is on the line.